This preview shows page 1. Sign up to view the full content.
Unformatted text preview: of different systems, we have learned some important lessons: ¯ ¯ Every system is different. Details about the hardware, operating system, and library function implementations can have a signiﬁcant effect on what kinds of programs can be measured and with what accuracy. Experiments can be quite revealing. We gained a great deal of insight into the operating system scheduler running simple experiments to generate activity traces. This led to the compensation scheme that greatly improves accuracy on a lightly loaded Linux system. Given the variations from one system to the next, and even from one release of the OS kernel to the next, it is important to be able to analyze and understand the many aspects of a system that affect its performance. Getting accurate timings on heavily loaded systems is especially difﬁcult. Most systems researchers do all of their measurements on dedicated benchmark systems. They often run the system with many OS and networking features disabled to reduce sources of unpredictable activity. Unfortunately, ordinary programmers to...
View Full Document
This note was uploaded on 09/02/2010 for the course ELECTRICAL 360 taught by Professor Schultz during the Spring '10 term at BYU.
- Spring '10
- The American